REDIAL-2020

REDIAL-2020: Ensemble Machine Learning for Anti-SARS-CoV-2 Activity Prediction

REDIAL-2020 predicts anti-SARS-CoV-2 activities using ensemble machine learning models trained on high throughput screening data from the NCATS COVID19 portal, estimating compound effects on viral entry, viral replication, and live virus infectivity.

Key Features:

• Machine Learning Models: Implements multiple categorical classifiers based on fingerprint, physicochemical, and pharmacophore molecular descriptors for parallel model development.
• Ensemble Consensus Predictor: Combines best-performing models into a consensus ensemble to improve predictive performance over individual classifiers, achieving external predictivity of 60–74% across three datasets.
• Multi-Target Activity Prediction: Estimates activities across viral entry, viral replication, and live virus infectivity using six independent models.
• Flexible Molecular Input: Accepts drug names, PubChem Compound IDs (CIDs), and Simplified Molecular Input Line Entry System (SMILES) strings for activity estimation.
• Similarity Search: Identifies structurally similar molecules based on experimentally determined activity data.

Scientific Applications:

• Drug Repurposing for SARS-CoV-2: Screens existing compounds to identify candidates with predicted anti-SARS-CoV-2 activity across multiple stages of viral interaction.

Methodology:

Models were trained on high throughput screening datasets from the NCATS COVID19 portal using multiple categorical machine learning algorithms and molecular descriptors (fingerprint, physicochemical, pharmacophore). An ensemble consensus approach integrates top-performing classifiers to optimize external validation performance.